We describe a tiled architecture that can fluidly transition between manycore (MIMD) and vector (SIMD) execution. The hardware provides a software-defined vector programming model that lets applications aggregate groups of manycore tiles into logical vector engines. In manycore mode, the machine behaves as a standard parallel processor. In vector mode, groups of tiles repurpose their functional units as vector execution lanes and scratchpads as vector memory banks. The key mechanism is an instruction forwarding network: a single tile fetches instructions and sends them to other trailing cores. Most cores disable their frontends and instruction caches, so vector groups amortize the intrinsic hardware costs of von Neumann control. Vector groups also use a decoupled access/execute scheme to centralize their memory requests and issue coalesced, wide loads.

This paper is motivated by work on a real-time Android navigation system using accelerometer, gyroscope, and camera data. We observed that our attempts at combining data from these different sensors yielded time series with jitter, i.e., inaccurate timestamps, that negatively affected program correctness and reliability. Thus, the aim of this paper is to help Android developers anticipate, assess, and understand problems with data quality in sensor-dependent applications. Our analysis focuses on jitter in timestamps obtained from Android's SensorManager and CameraManager services. We demonstrate the presence and severity of jitter obtained through these services and explain the characteristics of Android that originate jitter, resulting in a taxonomy of the various sources of jitter. Finally, we discuss the results of a quantitative analysis of jitter performed on three Android mobile devices (HTC One M8, Google Nexus 5, and Asus Zenfone 2) and provide recommendations on how Android developers can design sensing applications that will behave correctly across the majority of mobile devices.